Predictive Maintenance & Condition Monitoring - A Hot Seat Q&A Session

hello everybody and welcome to another

webinar from uh from exen Pro um my name

Gavin Green I look after strategic

solutions for exen pro and today I've

got Timothy White uh one of our

engineering Consultants uh with us today

so what are we going to cover into

today's session uh let's just build this

up so condition monitoring predictive

maintenance um essential for optimizing

equipment performance

reducing cost improving safety when you

start bringing in uh digital twins into

that um you can amplify some of these

benefits today's format of our webinar

is going to be a little bit different uh

not just us presenting information uh to

to the users um and demos Etc what we're

actually going to do is we're going to

do going to go through a few discussion

points specifically around condition

monitoring predictive maintenance uh

from an engineer's perspective um so

let's Dive Right In to that Tim do you

want to just give us a quick intro to

yourself please sure thanks Kevin uh

yeah I um I've done three years as a

reliability engineer in both process and

open pit and now I work as a software

consultant engineering consultant for XM

Pro perfect and also uh heavily involved

and and leading different projects so

Weir nutrient or a few of them as well

um and very active and he keeps pressing

us around process Improvement Etc so uh

very very very big advocate for that

Insight some of the topics that we're

going to cover um so fundamentals of

condition monitoring predictive uh we

have to start with a baseline where are

we coming from and then move from there

we're then going to go into Data

integration uh and

simulation um bringing in some realtime

data and iot integration we then going

to touch on some imp mation challenges

uh some of the data analytics and then

some best practices in predictive

maintenance um as well before we go into

these topic areas though um it's always

good to to just make sure that we have

some uh alignment on ter terminology so

we're all talking the same and we're not

talking past each other so just some of

the the common terms so when we're

talking around condition monitoring

we're talking around the the process to

continuously Monitor and assess perform

perance and health of Machinery um to

detect issues and prevent unplanned

downtime when we're talking about

predictive maintenance uh we're talking

about a proactive approach using data

analytics machine learning to predict

equipment failures so that we can

perform timely maintenance uh of them as

well and then the third prescriptive

maintenance um again using data

analytics machine learning this one

recommends specific actions and optimal

timing for maintenance tasks um to to

maximize reliability and minimize

downtime the terminology around a

digital twin So exm Pro is part of the

digital twin Consortium um the

definition uh from the Consortium does

not include the modelbased uh item that

we've got there um however we feel that

is uh it is part of the definition so

when we're talking digital twins we're

talking around a model-based virtual

representation of World entities and

processes synchronized at a specific

frequency and Fidelity we're going to

talk around digital twins as we go

through and and how that links into

condition monitoring predictive Etc but

just to make sure we're all on the same

Baseline what is it that we're talking

around and um the uh the common

definitions to to build from um as

well so let us move straight in

um to the first topic so so we're going

to put you in the hot seat uh hot seat

here Tim so to speak so the first area

that we're going to chat around is the

fundamentals of condition monitoring and

predictive maintenance so I've got a few

questions on the side here um we're not

going to bring the questions up be happy

to share with them after the webinar as

well so the first one companies looking

to to implement uh predictive

maintenance solution they often tell me

they're not ready for it um they don't

know where to start this is probably not

a fit for them how do you approach these

types of discussions with them that's uh

a really good question you have to

convince people also to spend money and

open the pur strings um so in in

reliability there's uh there's a

maintenance hierarchy that's typically

based on program maturity um we start

out at reactive preventative condition

based and predictive and prescriptive

you kind of touched on it a little bit

earlier uh I would say that company that

tries to do everything all at once is

going to waste quite a bit of money and

ruin any Goodwill that the reliability

group has generated with maintenance and

operations uh nobody starts with

prescriptive maintenance without quite a

bit of pre-work they start small with

condition-based monitoring and then move

into a PDM program um following that

maintenance hierarchy I would advise

this company to start with CD CBM or

condition-based monitoring and then move

to prescriptive and predictive modeling

uh as applicable based on asset

criticality okay okay now I do like that

um the fact that there is a hierarchy

there um and there there's there's steps

to to get to it so people don't just

jump into predictor maintenance from the

GetGo with nothing there's a few puddles

as we call them to to go through to to

actually get them um turning a little

bit though um how does condition

monitoring and predictive maintenance

benefit from bringing in and integrating

digital Twins and event intelligent

platforms into

that um that is a good question I I'll

start out with what uh an event

intelligence platform is um an event

intelligence

platform uh helps analyze data collected

from a variety of sources including

digital twins uh they help identify

patterns Trends uh anomal

IES um they can significantly help PDM

and condition monitoring reliability

groups by giving uh near real-time

alerts um to operation

staff uh using machine learning models

to detect deviations of sensors and

processes um they call it decision

support uh prescriptive

analytics um what that means is when a

certain predefined event criteria is

generated a uh alert or warning with

specific instructions on how to fix the

issue or mitigate the issue is also

generated in response to that uh they're

also incredibly useful for historical

data analysis

um I guess moving on to digital twins uh

they ideally ref reflect the current

condition of the physical

counterart uh this integration allows

for reliability teams to uh give

maintenance groups the ability to

visualize the uh they call it the life

cycle of a system and a controlled

environment um meaning from install to

failure they can model it with enough

data um reliability Engineers on that

note uh they can predict failures using

real-time data uh there's there's a

concept in reliability called the

potential for failure curve um based on

St statistical modeling you can put a

particular asset uh at a likelihood of

failure um planners can optimize

maintenance schedules basing rotation

off of condition rather than just time

and then uh on the more mature side data

scientists can test uh input scenarios

and improve designs through machine

learning

simulations um integrating digital Twins

and inet intelligence the I the most

important part is the the learning curve

is significantly shortened um you you

codify the uh experienced operator

intuition and

knowledge okay quite a bit to unpack

there um if we move into Data

integration and and

simulation so we've we've gone

through how do we get to condition

monitoring or predictive or prescriptive

of um the role the digital twin can

bring into that why it's

beneficial for all this to work

though data integration and simulation

going to play a key piece so what role

does data integration and

simulation play in condition monitoring

and predictive

maintenance uh a big one

so yeah uh data integration and

simulations play I call them crucial

roles in predictive maintenance and

condition monitoring uh especially

combined with digital Twins and event

intelligence platforms it's it's the

building

block uh focusing on data integration

first um just to make sure everyone's on

the same page it refers to the process

of combining data from multiple sources

into a single unified view um

consolidation of historical and

real-time data means merging historical

data with operational data patterns and

Trends can be more accurately

identified um it's it's your work order

history it's your shift notes it's your

vibration data in one

display um ideally with uh with regards

to simulation for condition monitoring

it uses computational models and machine

learning and and the more advanced

programs they they de to Ai and tensor

flow and things like that but it the

with the goal the main goal is to

imitate operations of real or processes

and systems um simulation allows for

testing the predictive model and uh

ensuring they're accurate before

applying them to actual equipment and

operations um lowrisk testing for V uh

variables scenario planning also helps

understanding how different conditions

might affect the system um game theing

better preparation and response

strategies uh lastly simulation provides

uh a virtual environment for training

operators and

technicians uh you you remove the risk

of actually damaging equipment and also

giving the operator an intuition without

that

risk together I guess data integration

and simulation they support creation of

a holistic

all-encompassing proactive maintenance

strategy um better decision- making they

provide detailed insights into equipment

Health under various conditions um

thereby I guess the the end result is

reducing downtime and extending life of

uh the

asset so is it also safe to say that uh

um from a data integration perspective

by combining all these different data

sources together um it allows for more

complex analysis which helps the

accuracy of these predicted results so

these two tend to work hand in hand yeah

yeah absolutely it's it's the building

block perfect perfect okay um so with

that being said though and the

criticality of the two of these and and

how they all work together um what

challenges exist in integrating some of

these data

sources uh quite a few um I guess one of

the biggest ones is uh call them data

silos so um different departments

different

organizations uh they have their own

systems and ways of doing things that uh

don't communicate very well with each

other and they can hinder that uh

unified view of data that a digital twin

requires um for example uh truck liner

bed thicknesses are on an Excel document

on one computer and fuel records are on

an Excel document and another computer

um both are useful to digital twins but

they aren't useful in their card format

and need to be exposed to the digital

twin um data quality and consistency is

another big one uh ensuring data is

accurate and up to dat and consistent

across the various sources is crucial um

because you're relying on that for

decision support poor data quality can

lead to incorrect analysis and faulty

predictions um complexity of integration

the data silos kind of touched on it but

um multiple data sources in Legacy

systems typically don't talk very well

with each

other um real-time data processing it's

it's costly it's expensive in in time

and money uh the ability to process

realtime data is challenging um but it's

also essential for timely decision

making for your PM groups operations

groups um scalability is another big one

uh the system must be able to scale with

growth and data and simulations without

also growing in latency nobody wants to

work on a slow system

um models in simulation

require

uh computational resources and the the

better the model uh or the excuse me the

I guess um models require High Fidelity

so the the better models require more

data and that's challenging without

continuous refinement and validation

outcomes and I touched on computational

resources the the more simulations you

run

um the more computationally intensive it

gets um cyber security is a big one as

we've all seen uh you're putting all of

your data in one bread basket and making

it a much bigger Target so cyber

security for everyone working is

incredibly

important um expertise in in digital

Twins and in reliability you you have to

be able to speak across departments for

both both the data scientists and the uh

reliability

folks um and we spoke on cost change

management is also a big one uh

workforces are generally resistant

anytime you have a technology like

digital Twins or event intelligence

machine learning um how the how the

software is rolled out and who's

involved and who has a stake in it can

make or break the program

um and then Regulatory Compliance as

well uh ensuring data handling and

processing comply with relevant

regulations whether it be socks or some

other regulation that I'm not aware of

um especially in industries that are

heavily

regulated uh addressing challenges

addressing the challenges requires

combination of strategic planning and

investment in technology and training

nobody's just going to happen upon it

and then development of new processes

and government or governance models for

data okay so there quite a few there's

quite a few there to unpack um if we if

we just focus on the the realtime um

aspect of this so real-time data because

you you alluded to it earlier around um

passing real-time data to the models uh

Etc so in in your experience what are

the biggest challenges and

opportunities when inter integrating iot

devices with intelligent digital twins

um specifically for realtime monitoring

in predictive maintenance condition

monitoring

Etc uh integrating iot devices presents

quite a few challenges into the um

digital twin uh one of the biggest ones

is data volume and management every iot

sensor you have um depending on the

uptake um generates quite a bit of data

and managing it can be it it needs

expertise if it everybody's dealt with a

system that isn't designed well um we've

talked about complex uh complexity of

integration but it's important so I'm

gonna repeat it uh it's iot systems

generally don't talk to each other so um

you need somebody that can go back and

forth between uh sensor compatibility

quality and connectivity is huge uh that

an accurate and expensive iot sensor

doesn't function without a reliable and

secure wireless network and guess the

other side of the coin

is uh there's no point to having a

reliable and secure network if the data

you're sending over it is inconsistent

and not

reliable interoperability of sensor

systems is big does your vibration

monitor also allow you to integrate uh

temperature

sensor

um regarding the change management side

of it and hesitancy to adopt new

technologies kind of along the same vein

nobody wants to go to five different

websites to interact with five different

sensor

types um I'm sure everybody's been in

that situation too uh lat latency and

skill of ility we we we talked about

latency a little bit but uh we need

asset views to load quickly regardless

of how many sensors or similar systems

exist and then maintenance um

maintenance of the digital twin is

crucial uh depending on size and

complexity they normally require subject

matter experts to maintain and modify

the uh hmis or human machine interfaces

but just like any other system you have

to maintain it uh that being said the

the rewards for a good integration are I

like to think they're exponential um the

pre-work done for one required for One

sensor is the same pre-work design for

20

sensors um data silos are somewhat

mitigated using

pipelines um exposing data making it

more easily

digestible uh it improves asset

performance through that proactive

maintenance strategy we talked about

prescriptive maintenance and then uh

economies of scale um the fixed cost for

sensor implementation becomes much more

palatable the more sensors you have

attack so that fixed cost per goes

down um and then transparency is Big so

all the stakeholders whether it be it or

reliability or maintenance or operations

are all working off the a common set of

data and then a short and learning curve

so the uh optimal State Intuition or

plant intuition the 20-year plant

operator has um can be codified into

that digital twin and that's why it's so

important to have that buy in from these

operators and mechanics that have been

it since I was in college um but yeah uh

another reward is automated decision

making um with the right integration

systems can automatically adjust

processes uh in response to data from

iot devices without hum uh human

intervention and then um safety

it it's a iot censor can exist in an

environment where a person can't U so

there absolutely

necessary um operational efficiencies

are improved when uh equipment is

operating at optimal levels everything

likes to run in steady state and that's

aided by iot

um addressing the challenges involves uh

crossplatform mix of technological

solutions strategic planning and ongoing

management it's not fire and forget

um yeah and then opportunities but

interesting um can you share a success

story uh where iot

integration um with a digital twin has

significantly

improved um either condition monitoring

or predictive maintenance

outcome uh yeah it and this also goes to

show how easily this stuff can be

integrated into your system um so a

while back

when uh I was working in

reliability we had a cone crusher that

would get packed in during Implement

weather uh the cone would bounce um once

that dirt gets packed in it becomes

basically cement and as the crusher

spins around it hops and if you can

imagine uh a 50 ton block of iron

hopping the uh wear and tear on it was

pretty tremendous it would sh shake the

entire building frame so obviously that

was not a desired outcome um the

vibration monitors on the cone crusher

were internal to it um so they could be

changed unless the cone crusher was shut

down so a operator came to me with the

idea to um use a Raspberry Pi and a

microphone

to

uh create a decel meter that wrote to

our

historian um so with a little bit of

python code and a Wi-Fi signal we we

created a proof of concept sensor that

uh fed directly into the control room

HMI and it worked in um backing up the

vibration monitors and sometimes um

replacing them whenever they went down

so um yeah worked out well and it was

cheap

uh cheap cheap uh in the grander scheme

of things yeah cheap in at

work okay so you've touched on some of

uh some of the items there um focusing

on implementation and and data um

analytics um what are some of the

implementation challeng implementation

challenges um of a predictive uh

maintenance solution when involving

digital

terms uh sure the the implement

challenges um for a digital twin for PDM

um they're call it multifaceted uh they

involve Technical Resources from it they

involve advice from reliability groups

operational resources and like I was

talking about those 20-year mechanics

and operators to provide advice in that

Common Sense

check um integrating with existing

systems we we sort of touched on on the

data stream designer earlier but this is

where it proves its worth uh for that

cone crusher example I was able to

perform the task without a lot of

pre-work because uh someone had built a

python library to access our historian

and

rewrite um without that library and

connection to the API it would have been

pretty much impossible at my current

skill level um the data stream designer

has quite a few options for interacting

with uh historians apis databases um

makes the barrier to entry much lower

when the API already

written um another uh challenge is data

collection and quality uh your model is

only as good as the sensors and the

people who help create it um some assets

need sophisticated sensors and some

don't take for instance a a redundant

pump you don't need to throw prescripted

maintenance and machine learning models

and everything like that if it's a $100

replacement and it's got a

redundancy um so you have to weigh asset

criticality as well

as ability and then uh it's this is a

this is a big one it delves into change

management a little bit it's absolutely

necessary that your IT personnel and

your data scientists at least have uh a

rudimentary understanding of the goals

behind a PDM digital twin and uh the

other side of that coin is your PDM

group has to be able to communicate

their needs to the data scientist team

the IT

team um model development validation and

maintenance is also another big

challenge um like I said before digital

twins aren't fire and

forget uh in addition to the regular

maintenance required by any software you

also have to consider business needs of

the plant change over time and the

digital twin has to evolve and grow uh

with the

operation oh go ahead you you touched on

it earlier and you touched on a little

bit here as well um so data analytics

and machine

learning how would they enhance some of

the predictive

capabilities uh data analytics and

machine learning for the more mature

programs um they are designed inherently

to detect patterns in data

sets um whether through supervised

learning or unsupervised

learning pattern recognition is what

they were designed for finding hidden

correlations um they also help in

anomaly of abnormal operating

parameters um predictive

maintenance uh excuse me predicted

modeling um what happens if I change X

and How likely is it going to be to

happen

um for optimizing maintenance schedules

and life cycle management um for example

you take a truck engine or a hall truck

engine uh life cycle can be 20,000 hours

or 40,000 hours depending on condition

and when you're

talking um a million and a half $2

million for an engine replacement it can

definitely affect the bottom line of a

company changing on condition rather

than

time root cause analysis is another one

um aided by anomaly detection and then

the findings of the root cause can then

be fed back into the digital twin to uh

enhance the prescriptive aspect of it

okay and enance decision making and

resource allocation

um clear analysis of data helps support

decision making and prioritizing

maintenance

activities perfect moving on to our last

uh section in our uh our last question

here just keeping an our time for the

folks as well is can you highlight some

best practices um in condition

monitoring and predictive

maintenance sure um some of the best

practices involve you you have to have a

strategy to start off with um clear

goals clear

objectives uh also you have to select

the appropriate

assets not every asset needs

prescriptive analytics like I was

talking about um integrate data

sources operational maintenance records

sensor data and environmental data were

available uh invest in quality

sensors use data analytics and machine

learning where applicable to helped prct

failure um ensure real-time

monitoring regularly update the machine

learning models to reflect current state

uh train staff create a feedback

loop um and most importantly communicate

effectively across departments for um

visibility

sake okay you you touched on a few

interesting ones there um so from an XM

Pro perspective um we follow some of

those best practices ourselves so how do

we how do we actually do that um the

first You' mentioned so identifying and

prioritizing Bad actors two predicting

in real time using a hybrid approach so

how do we bring in some of the you've

touched on condition monitoring

predictive prescriptive so how from a

hybrid approach can you bring some of

those pieces in there as well and then

the last piece is uh a quick time to

Value um for this um and that's aided

with our blueprints templates um Etc as

as

well I'd like to thank you uh thank you

Tim thank you for the time running

through this being being sitting in the

hot seat there so to speak U answering

the questions and things uh put to you

thank you all for uh listening and

attending as well um our our next

webinar next month um again there's two

options for you to attend uh from timing

perspective uh what we're going to do is

go through uh root cause analysis

application so we actually have a

blueprint for that as well how you

capture the recommendations value impact

Etc and as always if you've got any

questions feedback um we just want to

contact us for more information right at

the bottom there just send us an email

and we'll be happy to uh uh to provide

whatever it is uh you're looking for so

again Tim thank you for your time uh and

everyone else thank you for uh attending

today